Accepted Manuscript Title: Spectroscopic Analyses of Chemical Adaptation Processes within Microalgal Biomass in Response to Changing Environments Author: Frank Vogt Lauren White PII: DOI: Reference:

S0003-2670(15)00159-2 http://dx.doi.org/doi:10.1016/j.aca.2015.02.005 ACA 233715

To appear in:

Analytica Chimica Acta

Please cite this article as: Frank Vogt, Lauren White, Spectroscopic Analyses of Chemical Adaptation Processes within Microalgal Biomass in Response to Changing Environments, Analytica Chimica Acta http://dx.doi.org/10.1016/j.aca.2015.02.005 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Spectroscopic Analyses of Chemical Adaptation Processes within Microalgal Biomass in Response to Changing Environments Frank Vogt*, Lauren White

Department of Chemistry University of Tennessee 552 Buehler Hall Knoxville, TN 37996-1600 U.S.A. *corresponding author: [email protected] Graphical abstract

Highlights 

Microalgae transform large quantities of inorganics into biomass



Microalgae interact with their growing environment and adapt their chemical composition



Sequestration capabilities are dependent on cells’ chemical environments



We develop a chemometric hard-modeling to describe these chemical adaptation dynamics



This methodology will enable studies of microalgal compound sequestration



Abstract Via photosynthesis, marine phytoplankton transforms large quantities of inorganic compounds into biomass. This has considerable environmental impacts as microalgae contribute for instance to counter-balancing anthropogenic releases of the greenhouse gas CO2. On the other hand, high concentrations of nitrogen compounds in an ecosystem can lead to harmful algae blooms. In previous investigations it was found that the chemical composition of microalgal biomass is strongly dependent on the nutrient availability. Therefore, it is expected that algae’s sequestration capabilities and productivity are also determined by the cells’ chemical environments. For investigating this hypothesis, novel analytical methodologies are required which are capable of monitoring live cells exposed to chemically shifting environments followed by chemometric modeling of their chemical adaptation dynamics. FTIR-ATR experiments have been developed for acquiring spectroscopic time series of live Dunaliella parva cultures adapting to different nutrient situations. Comparing experimental data from acclimated cultures to those exposed to a chemically shifted nutrient situation reveals insights in which analyte groups participate in modifications of microalgal biomass and on what time scales. For a chemometric

description of these processes, a data model has been deduced which explains the chemical adaptation dynamics explicitly rather than empirically. First results show that this approach is feasible and derives information about the chemical biomass adaptations. Future investigations will utilize these instrumental and chemometric methodologies for quantitative investigations of the relation between chemical environments and microalgal sequestration capabilities. Keywords: Microalgal biomass, chemical adaptation, impacts of chemical environments, Fourier-Transform infrared, attenuated total reflection spectroscopy, chemometric ‘hardmodeling’

Introduction With an increasing industrialization, the production of anthropogenic CO2 is rising [ 1 ] and the fate of this greenhouse gas has become a serious concern [ 2 ], [ 3 ]. On the other hand, one major sink of atmospheric CO2 is photosynthesis and since marine phytoplankton is responsible for ~50% of the global primary production [ 4 ] - [ 10 ], it considerably contributes to counter-balancing anthropogenic CO2 releases [ 11 ] - [ 13 ]. In previous studies, the amount of produced microalgal biomass [ 14 ] - [ 16 ] and its chemical composition [ 17 ] - [ 20 ] have been linked to nutrient availability (C [ 21 ], N, P, Fe, and S) [ 22 ]. Apparently, the cells’ chemical nutrient utilization shifts when their chemical environment changes. For an accurate qualitative and quantitative assessment of a microalgal community’s sequestration of CO2 and other inorganic nutrients into biomass, an understanding of phytoplankton’s acclimatization mechanisms is crucial. Gathering such information on a culture-level can then contribute to a more accurate evaluation of the oceans’ carbon storage capacities which show first signs of saturation [ 23 ]. To investigate environmental impacts on phytoplankton biomass, HPLC [ 24 ] [ 26 ] and FTIR spectroscopy [ 17 ], [ 27 ] - [ 32 ] have been utilized; the latter technique in particular has gained popularity due to its sensitivity and selectivity for numerous, biologically relevant compounds [ 33 ]. The fact that nutrient levels and the presence of nutrient competitors are reflected in the biomass’ IR-spectroscopic signatures has been utilized in refs. [ 18 ], [ 19 ] for indirect analyses of growing environments. Another study [ 34 ] developed a novel chemometric methodology for FTIR-based quantitation of selected compounds within phytoplankton biomass. In these experiments, however, the biomass had to be fixed and thus only facilitated discontinuous analyses of the biomass’ chemical composition. Moreover, the algae’s chemical composition may be falsified during the biomass fixation as in particular volatile compounds may be lost during the drying step. Thus, for monitoring phytoplankton’s adaptation to a chemical shift in its growing environment, a novel approach is required that facilitates studying live cells. The required aqueous environment in turn hampers the use of FTIR transmission spectroscopy. In order to utilize the proven sensitivity and selectivity of FTIRspectroscopy, this study employed FTIR in attenuated total reflection (ATR) mode.

Experimental Cultures of the sea water species Dunaliella parva (The Culture Collection of Algae at the University of Texas at Austin) were grown in ‘semi-continuous mode’ [ 35 ] for eight days while being exposed to continuous illumination and 20ºC [ 14 ], [ 18 ]. The

cells were cultured in standard ‘Enriched Seawater, Artificial Water’ (ESAW) medium [ 35 ], [ 36 ] which contains natural nutrient concentration levels namely 2071µM HCO3and 549µM NO3-. These acclimated cultures were then transferred into a horizontal ATR accessory (PIKE Technologies) with a custom made ~100mL liquid cell placed on top of it. For the following 24h, FTIR-ATR spectra (4cm-1 resolution, 128 co-added scans) were recorded in 4min intervals with a Bruker Vertex 70 equipped with a DTGS detector. For subsequent chemometric studies, the wavenumber range 1350cm-1 – 950cm-1 had been chosen since many biologically relevant compounds namely sugars feature a distinct IR signature in this region [ 33 ]. During this 24h period, the microalgae cells slowly sink to the bottom (Fig. 1(a)) of the horizontal ATR cuvette where they are then probed by the mid-IR evanescent field. During this sinking process, the cells adapt to their chemical (nutrient) environment. Therefore, two processes determine the time dependent spectroscopic signatures: the accumulation of biomass within the reach of the evanescent field and changes in the cells’ chemical composition. In order separate both influences on the spectra, two types of experiments have been conducted: (i) Phytoplankton cultures were spectroscopically monitored without modification of their nutrient situation which thereby established a benchmark. The second type of experiment considered (ii-a) dilution of the nitrate concentration from 549µM to 419µM which has been achieved by adding 25mL of 160µM nitrate to 50mL of standard ESAW (549µM NO3-) in which the cells are contained. The other nutrients in the 25mL top-off were kept at the standard ESAW level and thus no dilution was introduced. (ii-b) Within other cell cultures, the nitrate concentration had been increased from 549µM to 793µM, 856µM, and 916µM, respectively, by adding 25mL of a higher concentrated ESAW to the 50mL standard ESAW. All cultures were prepared and analyzed in triple replicates. Fig. 1(b) depicts a time series of FTIR-ATR spectra acquired from a microalgae culture after spiking it to an elevated NO3- concentration. The increase of absorbance over time is clearly visible and a closer look reveals that in the beginning of the time series the absorption somewhat decreased. These changes in absorbances are hypothesized to be due to bio-sediment build-up and modification of the biomass’ chemical composition.

Theory Dynamics

Modeling Biomass’ Physical and Chemical

In order to explain such spectroscopic time series, a model function Y (v , t )  A(t )   (v , t ) for the measured data has been deduced which comprises of two factors, i.e. a time-dependent term A(t) which describes the accumulation of biomass within the reach of the evanescent field and a term,  (v, t ) , which describes wavenumber and time-dependent spectroscopic features. While only  (v, t ) is of immediate chemical and biological interest, the introduction of A(t) is required to properly explain the measured spectra. This model function Y (v , t ) is loosely related to Beer’s Law by interpreting A(t) as the biomass’ “concentration” which increases over time and  (v, t ) as wavenumber- and time-dependent “molar absorptivity”. Furthermore, as shown below, Y (v , t ) contains numerous model parameters  which reflect properties of a specific sample and will be extracted from experimental data via nonlinear least-squares regression. The long-term goal is to relate the values of the  s to ambient chemical

parameters and thus to explicitly link the chemical environment to the microalgae’s responses. Since Y (v , t ) is derived from theoretical considerations rather than in an empirically fashion, these  will carry interpretable spectrochemical and/or physical meanings based on which the key players of phytoplankton’s adaptation mechanisms can be studied. Hence, a chemometrics ‘hard-modeling’ approach has been chosen here over a more conventional ‘soft-modeling’ method [ 37 ], [ 38 ]. Modeling the Accumu lation of B io mass A(t) The following theoretical considerations about accumulating bio-sediment on top of the ATR element will lead to an ordinary differential equation (ODE) whose solution is A(t): 

Two mechanisms contribute to bio-sediment formation: (i) If gravitation exceeds buoyancy, cells sink and settle on top of the ATR crystal. (ii) Since the cells which are already within the reach of the evanescent field are alive, they keep multiplying.



Microalgae cells typically have a diameter (>5µm) which is in the same order of magnitude as the reach of the infrared ATR field. Thus, only the first layer of cells contributes to the ATR spectra. Once the first cell layer is completely covering the ATR element, the spectroscopic signal will not increase any further. For the subsequent discussion, A(t) describes the area of the ATR element covered by microalgae cells whereas the total area of the ATR element will be denoted by Amax  A(t ) .



The cells are assumed to be uniformly distributed in the culturing medium and thus every point of the ATR element is equally likely to be ‘hit’ by a sinking cell. Therefore, the increase in covered area A(t ) is proportional to the remaining free area, i.e.

dA  ( Amax  A(t )) . A currently unknown, time dependent function   t  completes dt this proportionality to a first-order ODE:

dA    t  .  Amax  A  t   dt (1)



  t  is assumed to be composed of three components: (i) The higher the sinking velocity vsin k of the cells, the faster the biofilm builds up. (ii) The larger cell size S cell , the more area is covered by each cell added on top of the ATR element. (iii) The higher the cell concentration cell  t  in the culturing medium, the more cells are reaching the ATR crystal and the faster the bio-sediment builds up. Therefore:   t   S cell  vsin k  cell  t  .



The cell size S cell is approximated to be constant and nutrient independent. In future investigations, size distributions [ 14 ] - [ 16 ] could be incorporated to replace a fixed cell size.



The sinking velocity vsin k is determined by the net force F=Fg-Fb between gravitation Fg and buoyancy Fb. Here, a constant buoyancy and gravity is assumed and thus vsink is only determined by the duration the acceleration a acts upon the cells:

vsin k  t    a dt  

F F dt   t  v0 with v0 denoting an unknown initial sinking mcell mcell

velocity. Yet, for this initial study, it has been assumed that the acceleration is small enough to justify the approximation vsin k  const . 

The viscosity and density of the culturing medium are assumed to be the same for all experiments and are incorporated into vsin k .



Since the cells are alive and provided with nutrients, they continue to divide and thus the cell concentration cell  t  in the culturing medium is governed by a sigmoidal function of time t [ 16 ]. In ( 2 ), max describes the maximum cell concentration sustained by a given culturing environment, 0 states the inoculated cell concentration, and s represents a culture’s growth rate:

cell  t   max 

1     max 1  exp    s   t  t0     max  1   max  0   0 (2)

After incorporating all information into ( 1 ), the resulting ODE

1 Amax  A

dA  Scell  vsin k  max 

1     max 1  exp    s   t  t0    max  1   max  0   0

dt

(3)

is solved via ‘separation of variables’ [ 39 ] and results in ( 4 ) with C being an integration constant [ 40 ]:

A  t   Amax

   0  max    C  1  max  exp   s   t  t0  0   max  0  

 Scell vsin k 

max  0 s

(4)

Equation ( 4 ) describes how, over time, the portion A(t) of the total ATR surface Amax is being covered by microalgae cells. For subsequent regression purposes, model parameters 1,,6 are introduced into ( 4 ) which are then determined from experimental data:







A  t   1   2  1   3  exp  4   t  5 

 6

(5)

M o d e l i n g C h a n g i n g S p e c t ro s c o p ic S i g n a t u r e s    ,t  The chemically adapting biomass gives raise to time dependent spectroscopic signals which are modeled by a function    ,t  . Generally, this    ,t  comprises of

four additive absorbance contributions namely a static term, a term describing the consumption of reactants, the formation and consumption of intermediates, and the formation of products:    , t    static      reac tan t  , t    int ermediate  , t    product  , t  . According to Beer’s Law, each of these  -terms is a product of an analyte’s molar absorptivity at a given wavenumber  and its concentration. While information about molar absorptivities reveal what compounds are present in a microalga, from concentration time-profiles, chemical adaptation dynamics can be deduced. However, explicitly modeling the time dependency of such chemical reactions is not straightforward because the reaction order is unknown and multiple reactions may occur at a given wavenumber  . In this first study, two simplifying assumptions are made: (i) only one analyte absorbs at  but different analytes can be present at different wavenumbers; (ii) chemical modification of the biomass occurs as two consecutive firstk1

k2

order reactions: R  I  P , and therefore the time-dependent concentrations of reactant R, intermediate I, and product P are given as [ 42 ], [ 43 ]:

 R t   R 0  exp   k1  t  I t   R 0  k

k1   exp   k1  t   exp   k 2  t   2  k1

(6)



 k2 k1  exp   k1  t    exp   k 2  t   k2  k1   2  k1 These concentration time-profiles ( 6 ) are multiplied with the unknown molar absorptivities a   of reactant, intermediate, and product, respectively. In order to extract relative concentrations and their temporal changes, wavenumber-dependent model parameters     are introduced whose numerical values will be determined from experimental data via nonlinear least-squares. Since at different wavenumbers different analytes contribute to the spectroscopic signals, the model parameters     are wavenumber dependent. Therefore:

 Pt   R 0 1  k

 reac tan t   , t   areac tan t     R t 

     exp   9    t     k1   areac tan t   R 0

8   

(7)

 int ermediate   , t   aint ermediate     I t 

10      aint ermediate     R 0 

k1 k2  k1

          exp   11     t   exp   12    t        k1     k2    (8)

 product   , t   a product     Pt          13     1  14     exp   15    t   16    exp   17    t             k1   k2  k2  k1  a product    R 0    k2  k1 k 2  k1 (9)

Furthermore, chemically stable compounds may be present within the biomass which contribute a time independent but wavenumber dependent term to the spectra. Their spectroscopic signature is described by ( 10 ) in which astatic    denotes the molar absorptivity of this mix and [S] its concentration.

 static      7   

 astatic    S 

( 10 )

Lastly, in order to incorporate a spectroscopic baseline drift, a constant spectroscopic offset, 18   , is added to Y   , t   A  t      , t   18   . In conclusion, the final model function Y   , t  to which spectra time series will be fitted is: Y (v , t )  1   2  {1  3  exp{ 4  (t  5 )}}6    7 (v )  8 (v )  exp{9 (v )  t}  10 (v )  [exp{11 (v )  t}  exp{12 (v )  t}  13 (v )  [1  14 (v )  exp{15 (v )  t}  16 (v )  exp{17 (v )  t}]]  18 (v ) ( 11 )

Since this model function ( 11 ) has been deduced by chemometric ‘hardmodeling’ [ 37 ], its model parameters 1,,18 Error! No bookmark name given. reflect chemical and physical properties (see ( 4 ) - ( 9 )). After determining their numerical values from time series of FTIR-ATR spectra (Fig. 1(b)), chemical processes can be analyzed that are initiated by a chemical shift in the microalgae’s growing environment. This will then enable more detailed insights into microalgae  ecosystem interactions. I m p l e m e n t i n g N o n l i n e a r L e a s t -S q u a r e s R e g r e s s i o n The model function Y   , t  ( 11 ) is highly nonlinear in the model parameters  and therefore the nonlinear regression algorithm is prone to derive a suboptimum fit solution ( ‘local minimum problem’). In this particular application, nonlinear regression can be employed via one of two different routes: (i) The Residual Sum of Squares (RSS) is minimized incorporating all N wavenumber positions into one regression. This is preferable since A  t  ( 5 ) is common to all wavenumbers and one common fit ensures wavenumber-consistent values for 1,,6 . However, the number of model parameters  is huge, i.e. 6  N 12  6  208 12  2502 , and the more model parameters are involved (i.e. the higher-dimensional the parameter space is), the more local minima exist. (ii) Therefore, fitting ( 11 ) wavenumber-by-wavenumber has been chosen here as this

involves N=208 fits comprising of only 6  112  18 nonlinear model parameters  each. While there are many software tools for performing nonlinear least-squares (e.g. [ 42 ] - [ 45 ]), the key to a good solution lies in providing the algorithm with an appropriate initial guess of the model parameters  . Therefore, Guided Random Search [ 46 ] has been utilized which estimates probability distribution functions (pdfs) which describe the likelihood that a given  leads to a low RSS and a high correlation coefficient. For applying Guided Random Search, boundaries for the  s are required which can straightforwardly be deduced since the  s possess known chemical meanings: Comparing ( 4 ) and ( 5 ) suggests to choose 0  1,2,4  1 , 0  3  1000 , 500  5  500 , and 0   6  0.2 to model the biomass accumulation. For setting boundaries for model parameters involved in describing the chemical adaptation, ( 7 ) ( 9 ) are assessed: First, the boundaries for 7 depend on whether a baseline drift caused negative absorbances or not. In the former situation, 0  7  y   , t   max yn 1,, N  min  yn  has been chosen with yn representing





measured data points. In absence of negative absorbances, 0   7  max  yn  , has been set and based on that, min  7   8,13  max  7  follows If no reactions occur, 8,10,13  0 is implied. Given the time scale, 0  9,11,12,15,17  0.1 has been chosen to reflect a reasonable reaction rate constants (= time constants). For 10,14,16 , positive and negative values are allowed since k2  k1 can be positive or negative; 100  10,14,16  100 was found empirically to be reasonable. Lastly, two situations are discriminated regarding 18 which models a potential spectroscopic baseline drift: If a negative absorbance y0  0 has been recorded at t  0 , a baseline drift has occurred for sure and requires modeling; thus, the following boundaries were chosen 1.1 y0  18  0.9  y0 . If however y0  0 , a potential baseline drift cannot be discriminated from signal-generating biomass already being within the reach of the evanescent field; therefore 18  0 is enforced. To reduce the risk of the iterative nonlinear least-squares becoming stuck in a local RSS-minimum, regression constraints are helpful. Based on the interpretability of the model function, additional information about this particular system is available which has been incorporated via equality and inequality constraints [ 34 ], [ 45 ]. Five equality constraints are applicable here (see( 11 )):

9    11    15   12    17   14   

12   12    11  

11   12    11   Furthermore, four inequality constraints were deduced: 16   





 6

1   2  1  3  exp  4   t  5  because the area of the ATR element covered at any time t cannot exceed the maximum area of that element. 10     exp  11    t   exp  12    t    0 because the contribution of the intermediate analyte to the spectra cannot be negative. 13    1  14    exp  15    t  16    exp  17    t    0 because the contribution of the product to the spectra cannot be negative. d 13     1  14    exp  15    t  16    exp  17    t    13    14    15    exp  15    t    dt because no product will be lost once it has been generated and therefore the signal contribution of the product cannot decline.





Results and Discussion The model function Y   , t  ( 11 ) has been fitted, one wavenumber at a time, to all data sets acquired from D. parva cultures grown under the stated nutrient scenarios. Fig. 1(c) and (d) depict representative fit outcomes at two different wavenumber positions of three replicate cultures which are in good agreement with the experimental data points. Panel (e) shows the time dependent spectroscopic profiles ( 7 ) - ( 10 ) for replicate #1 acquired at 1076cm-1 and panel (f) displays the accumulation of bio-sediment A  t  ( 5 ). Multiplying the dash-dotted curve shown in (e) with the bio-sediment coverage (f) plus the baseline offset 18    (not shown) results in the fit curve ‘replicate #1’ displayed in (c). Goal of this study is to determine whether the values of the model parameters 1,,17 ( 11 ) [ 48 ] are impacted by a change in the NO3- concentration. Due to the chemical and physical interpretability of these parameters, it can be assessed what type of adaptations these environmental chances trigger within the phytoplankton biomass. While 1,,6 mostly reflect physical parameters such as buoyancy, cell size, and growth characteristics,  7,,17 represent chemical information such as contribution of reactants, intermediate compounds, and products to the biomass’ spectroscopic signature as well as reaction rate constants (or more generally: time constants of chemical adaptations). As explained above, the fits have been performed one wavenumber at a time and thus N values for each 1 , ,  6 have been obtained per sample. In Fig. 2, these values have been compiled for all concentration scenarios considered here including their replicates. In order to estimate a sample’s true values for 1 ,,  6 , the fits’ values were averaged over all N wavenumber positions. Based on these mean values and their standard deviations, impacts of nutrient shifts on the cultures ‘physical parameters’ were assessed: With one outlier sample, 1  Amax is very reproducible across the different nutrient conditions. The fact that the ATR crystal’s surface was not found to be impacted by the cells’ nutrient conditions is reassuring that the nonlinear regression algorithm derives valid fits despite the potentially large number of local RSS-minima.  2 features the largest sample-tosample fluctuations which is not surprising since it incorporates the initial coverage Amin [ 40 ] of the ATR element. Variation of this parameter had been expected since manually

filling the ATR cuvette is not very reproducible. The mean values for 3 and 4 are very consistent between samples but feature large errorbars. This may be indication that a culture’s initial and final cell concentration are not very relevant to the accumulation of bio-sediments. At first this may be surprising but one has to keep in mind that only the first cell layer covering the ATR crystal contributes to the ATR spectra. Apparently the cultures had enough cells to produce this layer so that the cell growth only plays a minor role. 5 and 6 feature consistent means and comparatively small errorbars and do not indicate that there are impacts of nutrient shifts on cell size and the build-up of the biosediment. Based on these first results, it is concluded that the nutrient chance has no measureable impact on 1,,6 . Comparing model parameters  7,,17  v  from samples provided with a steadystate NO3- concentration to their counterparts derived under a changing nutrient availability provides insights into phytoplankton’s chemical adaptations. Considering the model parameters meaning, an assessment of the adaptation processes can then be made in relation to the nutrient concentration change. Since this study focuses on a proof-ofprinciple demonstration of monitoring live-cell adaptation processes, assessing which analytes within the biomass are involved has not been attempted here. Fig. 3 - Fig. 6 show impacts on the nutrient concentration change on 7,8,9,13  v  ; results for the remaining parameters were compiled into the Supplemental Material. In all graphs, parameters’ values for the steady-state (black) are compared to their counterparts computed after a nutrient change (red). The   values for three replicate cultures have been averaged wavenumber-by-wavenumber in order to incorporate natural fluctuations. To detect statistically significant differences,  nsteady  state (black) and  nchanged (red) were ttested (95% confidence) wavenumber-by-wavenumber. These t-tests’ outcomes are also included in Fig. 3 - Fig. 6: If at a given wavenumber position no significant difference between  nsteady  state and  nchanged was found, the t-test result is encoded (▲) by a zero. If the nutrient change resulted in  nsteady  state   nchanged , that wavenumber position was marked by a positive number for the ▲. If  nsteady  state   nchanged , a negative value was assigned to the corresponding ▲.  7  v    static   describes the phytoplankton’s spectroscopic features that do not change over time. However, this does not imply that all samples originating from identical growing conditions result in equivalent  7  v  : As has been discussed above, 1,,6 does not notably change with the nutrient situation and since the same spectroscopic equipment has been utilized in all experiments, it is reasonable to assume that the measured absorbances of all samples are –numerically- in the same range. If chemical adaptations occur within the biomass, it can be expected that the static spectroscopic signatures drop in absorbance when reactants, intermediates, and products are adding to the spectroscopic signatures. This expectation has been confirmed (Fig. 3) has clearly been confirmed for three out of four nutrient changes as the t-tests found that  7 values are significantly lower (negative values for ▲). For the concentration spike to 793µM, the errorbars were notably higher and thus a less clear indication of decreased  7

values was found. For smaller concentration changes, a reduction in static compounds was mostly found between 1350cm-1 and 1200cm-1; for the strongest nitrate spike, spectroscopic changes spread towards lower wavenumbers as well. Fig. 4 displays the impact of the changing nutrient concentration on the spectroscopic signature of the reactants 8  v  in the biomass. It had been expected that the steady-state nutrient scenario would lead to 8  0 which has not been confirmed. One explanation is that the cells continue to consume nutrients, which were not replenished over the course of these 24h experiments. Hence the cells continuously alter their own growing environment which in return may trigger the cells to chemically adapt. The additional nutrient change amplifies these adaptations as demonstrated by numerous▲≠0. It appears that the number of wavenumber positions at which the reactants’ spectroscopic signature changes is increasing with the disturbance’s strength. In this limited wavenumber range (1350 cm-1-950cm-1) there is a trend towards decreasing the reactants’ contribution to the spectra. The reactants may be more prominent in other wavenumber regions though. Changes of the reaction rate constant k1  9   are compiled in Fig. 5. Obviously, no clear wavenumber dependency among these values was found in this study. While the nutrient situation has a statistically significant impact on a few 9 s , the overall outcome does not strongly support that a trend in either increase or decrease of the reaction speed is measurable in this application. Lastly, the spectroscopic contribution of the products, i.e. 13    , are assessed by means of Fig. 6. Again, the initial expectation of no product formation in case of a steady-state nutrient scenario has not been confirmed. The 13  values shown in black are clearly ≠0. Whether product formation is due to continued, self-imposed environmental changes, i.e. nutrient consumption by the life cells, requires further investigations. Comparing the 13  values as obtained under steady-state nutrient concentrations to the dilutions or spiking scenarios reveals that the more the nutrient concentration deviates from the natural level of 549µM, the lower the products’ contribution to the spectra become. Whether there are no products formed that absorb in the chosen wavenumber region requires additional investigations. For the remaining model parameters, equivalent plots can be found in the Supplemental Material.

Conclusions and Outlook Microalgae play a major role in ecosystems as they sequester large quantities of the greenhouse gas CO2 and other inorganic nutrients out of the environment into biomass. Previous studies have demonstrated that chemical shifts in their growing environment initiate changes in the chemical composition of their biomass. For assessing whether this impacts microalgal compound transformation as well it is important to study, on a cell-culture level, how phytoplankton adapts to a changing ecosystem in which they are embedded. The microalgae species Dunaliella parva had been selected for such studies and exposed in laboratory experiments to shifting chemical environments. FTIR-ATR spectroscopy has been chosen for monitoring chemical signatures of live biomass based on this technique’s selectivity and sensitivity for many biologically relevant compounds.

For analyses of resulting spectroscopic data, a novel chemometric ‘hard-modeling’ approach has been developed with the goal to explicitly describe the chemical, physical, and some physiological processes the cells undergo during chemical adaptations. Of particular interest were the consumption of reactants as well as the formation of intermediate compounds and final products. Ultimate goal is the development of a methodology for extensive investigations of how environmental changes drive the chemical adaptation of biomass. With such an analytical tool at hand, assessments of marine ecosystems’ capabilities to sequester inorganic compounds into biomass will become feasible. First results presented here clearly demonstrate that FTIR-ATR is feasible for monitoring live cell adaptations and that the introduced model describes well the dynamic, statistically significant shifts in the biomass’ chemical composition. This study focused on developing a methodology which in subsequent investigations will be applied to gain chemical insights.

Acknowledgement This work was supported by the National Science Foundation under CHE1058695 and CHE-1112269. References [ 1 ] G. Peters, G. Marland, C. Le Quéré, T. Boden, J. Canadell, M. Raupach. Rapid growth in CO2 emissions after the 2008–2009 global financial crisis, Nature Climate Change 2 (2012) 2-4. [ 2 ] M. Eby, K. Zickfeld, A. Montenero, D. Archer, K. Meissner, A. Weaver. Lifetime of Anthropogenic Climate Change: Millennial Time Scales of Potential CO2 and Surface Temperature Perturbations, Journal of Climate 22 (2009) 2501-2511. [ 3 ] J. Blunden, D. Arndt, M. Baringer (Eds.). State of the Climate in 2010, Bulletin of the American Meteorological Society 92 (2011) S1–S266. [ 4 ] C. Field, M. Behrenfeld, J. Randerson, P. Falkowski. Primary production of the biosphere: integrating terrestrial and oceanic components, Science 281 (1998) 237-240. [ 5 ] M. Behrenfeld, R. O’Malley, D. Siegel, C. McClain, J. Sarmiento, G. Feldman, A. Milligan, P. Falkowski, R. Letelier, E. Boss. Climate-driven trends in contemporary ocean productivity, Nature 444 (2006) 752 – 755. [ 6 ] E. Martinez, D. Antoine, F. D’Ortenzio, B. Gentili. Climate-Driven Basin-Scale Decadal Oscillations of Oceanic Phytoplankton, Science 326 (2009) 1253-1256. [ 7 ] D. Bilanovic, A. Andargatchew, T. Kroeger, G. Shelef. Freshwater and marine microalgae sequestering of CO2 at different C and N concentrations – Response surface methodology analysis, Energy Conversion and Management 50 (2009) 262–267. [ 8 ] J. Raven, M. Giordano, J. Beardall, S. Maberly. Review: Algal and aquatic plant carbon concentrating mechanisms in relation to environmental change, Photosynthesis Res. 109 (2011) 281–296. [ 9 ] M. Giordano, J. Beardall, J. Raven. CO2 Concentrating Mechanisms in Algae: Mechanisms, Environmental Modulation, and Evolution, Ann. Rev. Plant Bio. 56 (2005) 99-131. [ 10 ] J. Raven, M. Giordano, J. Beardall, S. Maberly. Review: Algal evolution in relation to atmospheric CO2: carboxylases, carbon-concentrating mechanisms and carbon oxidation cycles, Phil. Trans. R. Soc. B 367 (2012) 493–507.

[ 11 ] G. Hays, A. Richardson, C. Robinson. Climate change and marine plankton, Trends in Ecology and Evolution 20(6) (2005), 337-344. [ 12 ] R. Bardgett, C. Freeman, N. Ostle. Mini-Review: Microbial contributions to climate change through carbon cycle feedbacks, The ISME Journal (International Society for Microbial Ecology) 2 (2008), 805–814 [ 13 ] J. Zhou, K. Xue, J. Xie, Y. Deng, L. Wu, X. Cheng, S. Fei, S. Deng, Z. He, J. Van Nostrand, Y. Luo. Microbial mediation of carbon-cycle feedbacks to climate warming, Nature Climate Change 2 (2012), 106-110 [ 14 ] M. McConico, R. Horton, K. Witt, F. Vogt. Monitoring chemical impacts on cell cultures by means of image analyses, J. Chemom. 26 (2012) 585–597. [ 15 ] M. McConico, F. Vogt. Modeling nutrient impacts on microalgae cells via image analyses, J. Chemom. 27 (2013) 217–219. [ 16 ] L. White, D. Martin, K. Witt, F. Vogt. Impacts of nutrient competition on microalgae biomass production, J. Chemom. 28 (2014) 448–461. [ 17 ]

M. Giordano, A. Domenighini, S. Ratti, F. Vogt, Spectroscopic Classification of 14 Different Microalgae Species – First Steps towards Spectroscopic Measurement of the Biodiversity in Marine Environments, Plant Ecology and Diversity, 2 (2009) 155-164.

[ 18 ] R. Horton, M. McConico, C. Landry, T. Tran, F. Vogt. Introducing Nonlinear, Multivariate ‘Predictor Surfaces’ for Quantitative Modeling of Chemical Systems with Higher-order, Coupled Predictor Variables , Anal. Chim. Acta 746 (2012) 114. [ 19 ] M. McConico, F. Vogt. Assessing Impacts of Nutrient Competition on the Chemical Composition of Individual Microalgae Species, Anal. Lett. 46 (2013) 2752–2766. [ 20 ] M. Giordano, S. Ratti. The biomass quality of algae used for CO2 sequestration is highly species-specific and may vary over time, Journal of Applied Phycology 25 (2013) 1431–1434. [ 21 ] Note: Sequestering atmospheric CO2 by phytoplankton in aqueous media occurs mostly by uptake of HCO3− [ 9 ] which is produced via: CO2 (g) ↔ CO2 (aq) + H2O ↔ H2CO3 ↔ HCO3− + H+ ↔ CO32− + 2 H+ [ 22 ] M. Behrenfeld, K. Halsey, A. Milligan, Review: Evolved Physiological Responses of Phytoplankton to their Integrated Growth Environment, Phil. Trans. Royal Soc. B 363(1504) (2008) 2687-2703. [ 23 ] C. Quéré, C. Rödenbeck, E. Buitenhuis, T. Conway, R. Langenfelds, A. Gomez, C. Labuschagne, M. Ramonet, T. Nakazawa, N. Metzl, N. Gillett, M. Heimann. Saturation of the Southern Ocean CO2 sink due to recent climate change, Science 316 (2007) 1735–1738. [ 24 ] J. Rodriquez, G. Reina, J. Rodriquez. Determination of free amino acids in microalgae by high-performance liquid chromatography using pre-column fluorescence derivatization, Biomed. Chromatogr. 11 (1997) 335-336. [ 25 ] M. Rigobello-Masini, J. Penteado, C. Liria, M. Miranda, J. Masini. Implementing stepwise solvent elution in sequential injection chromatography for fluorimetric determination of intracellular free amino acids in the microalgae Tetraselmis gracilis, Anal. Chim. Acta 628 (2008) 123-132.

[ 26 ] J. Penteado, M. Rigobello-Masini, C. Liria, M. Miranda, J. Masini. Fluorimetric determination of intra- and extracellular free amino acids in the microalgae Tetraselmis gracilis (Prasinophyceae) using monolithic column in reversed phase mode, J. Sep. Sci. 32 (2009) 2827-2834. [ 27 ] M. Giordano, M. Kansiz, P. Heraud, J. Beardall, B. Wood, D. McNaughton. Fourier Transform infrared spectroscopy as a novel tool to investigate changes in intracellular macromolecular pools in the marine microalga Chaetoceros muellerii (Bacillariophyceae), J. Phycol. 37 (2011) 271-279. [ 28 ] P. Heraud, B. Wood, M. Tobin, J. Beardall, D. McNaughton. Mapping of nutrientinduced biochemical changes in living algal cells using synchrotron infrared microspectroscopy, FEMS Microbiol. Lett. 249 (2005) 219-225. [ 29 ] K. Stehfest, J. Toepel, C. Wilhelm. The application of micro-FTIR spectroscopy to analyze nutrient stress-related changes in biomass composition of phytoplankton algae, Plant Physiol. Biochem. 43 (2005) 717-726. [ 30 ] C. Hirschmugl, B. Zuheir-El Bayarri, M. Bunta, J. Holt, M. Giordano. Analysis of nutritional status of algae by Fourier transform infrared chemical imaging, Infrared Phys. Techn. 49 (2006) 57-63. [ 31 ] A. Domenighini, M. Giordano. Fourier transform infrared spectroscopy of microalgae as a novel tool for biodiversity studies, species identification, and the assessment of water quality, J. Phycol. 45 (2009) 522-531.

[ 32 ] J. Murdock, D. Wetzel. FT-IR Microspectroscopy Enhances Biological and Ecological Analysis of Algae, Appl. Spec. Rev. 44 (2009) 335 – 361. [ 33 ] R. Horton, E. Duranty, M. McConico, F. Vogt, FTIR Spectroscopy and Improved Principal Component Regression for Quantification of Solid Analytes in Microalgae and Bacteria, Appl. Spec. 65 (2011) 442-453. [ 34 ] F. Vogt, Information fusion via constrained principal component regression for robust quantification with incomplete calibrations, Anal. Chim. Acta 797 (2013) 20– 29. [ 35 ] R. Andersen, Algae Culturing Techniques, Elsevier Academic Press, Burlington, 2005. [ 36 ] J. Berges, D. Franklin, Evolution of an artificial seawater medium: Improvements in enriched seawater, artificial water over the last two decades, Journal of Phycology 37 (2001) 1138 -1145. [ 37 ] F. Vogt. Quo vadis, chemometrics? J. Chemom. 28 (2014), 785-788 [ 38 ] Note: Multivariate Curve Resolution (MCR) would be an alternative that can extract spectroscopic and time profiles from experimental data and thus can deduce an interpretable model Y   , t  . However, MCR is only applicable to models such as Z   , t   B  t       in which the variables, here time and wavenumber, are separable. For the given experiment, however, the model function Y   , t   A  t     , t  contains a time-dependent spectroscopic term

   ,t  which couples time and wavenumber information and thus renders the model to be nonlinear.

[ 39 ] M. Tennenbaum, H. Pollard, Ordinary Differential Equations. Dover Publications: Mineola, NY, 1985. [ 40 ] Note: Solving the ODE ( 3 ) involves calculating an integral of type 1  1  exp   x dx  ln 1  exp  x [ 41 ] or more generally 1 1  1    exp      x    dx    ln 1    exp     x    . (Going from the





former to the latter integral requires a change of variable 1    xold     ln    xnew in conjunction with dxold   dxnew .) In the given    0 max case, these parameters are   max ,   s , and   t0 Utilizing 0 max  0 all the above considerations leads to:

 ln  Amax  A  S cell  vsin k 

   0  max  max  0  ln 1  max  exp   s   t  t0   C s 0   max  0 

and hence

A  Amax

     Scell vsin k . max 0   s  max    max  0    exp C  exp  ln 1   exp   s   t  t0      0   max  0      C    

A  t   Amax

 

   0  max   1  max  exp   s   t  t0   0   max  0 

 Scell vsin k 

max  0 s

C

In theory, the integration constant C contained in the above equation could be derived from the initial condition that A  t  0  Amin :

C

Amax  Amin  Scell vsin k v

max  0

s  max  0   max 1   exp   s  t   0  0  max  0   However, from an experimental perspective, determining in initial coverage Amin of the ATR element with biomass is challenging and here it is more straightforward to consider C   2 as one of the model parameters in ( 5 ). [ 41 ] M. Abramowitz, I. Stegun, Handbook of Mathematical Function with Formulae, Graphs, and Mathematical Tables (10th printing). Dover Publications: Mineola, NY, 1972 [ 42 ] E. Billo, Excel for Chemists – a Comprehensive Guide, 2nd ed., Wiley-VCH: New York, 2001

[ 43 ] Note: From a different perspective, exponential time dependencies are common in nature and serve as a generic description in which the (reaction rate) constants k1,2 determine time scales on which the chemical composition changes. [ 44 ] M. Galassi, J. Davies, J. Theiler, B. Gough, G. Jungman, P. Alken, M. Booth, F. Rossi, GNU Scientific Library Reference Manual (3rd ed., software V1.12). Network Theory Ltd, 2009.http://www.gnu.org/software/gsl/ accessed on: January 15, 2015 [ 45 ] http://ab-initio.mit.edu/wiki/index.php/NLopt accessed on: January 15, 2015 [ 46 ] F. Vogt, A self-guided search for good local minima of the sum-of-squared-error in nonlinear least squares regression, J. Chemom. 2014, early view online, DOI: 10.1002/cem.2662 [ 47 ] Note: In panel (d), absorption’s time profiles are depicted and not concentration time profiles; therefore the strong signature of the intermediate compound is not unrealistic since these compounds may feature large values for aint ermediate ( 8 ). [ 48 ] Note: 18  v  has been left out of this discussion as it only incorporates spectroscopic baseline drifts into Y   , t  to prevent fit distortions.

Figure captions



Fig. 1: Example of fit results for a sample after spiking the NO3 concentration at t=0 from the ambient level, i.e. 549µM , to 856µM ; the fact that the absorbances are negative can be attributed to baseline drifts which are incorporated into Y  , t by means of 18  ( 11 ). (a) accumulation of bio-sediment at the bottom of a cell culture; (b) time series of mid-IR spectra monitoring the build-up of chemically adapting bio-sediment; (c) time profile of the bio-sediment’s

 

measured and fitted absorption at 960cm-1 and (d) at 1076cm-1; (e)

 

 static    960cm 1  ( 10 );

 reac tan t  , t  ( 7 );  int ermediate  , t  ( 8 ),  product  , t  ( 9 ), and the total spectroscopic signal   ,t  [ 47 ]; (f) bio-sediment A(t) build-up ( 5 ); the profiles (d) and (e) were computed from the corresponding model parameters

Fig. 2: Analyzing the dependency of

1 ,,  6 on changes in the NO3 concentration in the



culturing medium; the final NO3 concentrations after diluting/spiking are stated on top of the replicates separated by vertical dashed lines



7 ( 10 ); a positive value for a t-test outcome (▲) indicates that the corresponding value for  7 Fig. 3: Impact of NO3 dilution or spiking on the time independent spectroscopic signature

increased due to the concentration change – a negative value for the t-test depicts a decreased value for  7



Fig. 4: Impact of NO3 dilution or spiking on the reactant’s spectroscopic signature at t  0 , i.e.

8 ( 7 ); see Fig. 3 for explanation of the t-test outcomes



Fig. 5: Impact of NO3 dilution or spiking on the reactant’s decay constant

9 ( 7 ) (or: reaction

rate constant k1 ) ; see Fig. 3 for explanation of the t-test outcomes



Fig. 6: Impact of NO3 dilution or spiking on the product’s spectroscopic signature at t  0 , i.e.

13 ( 9 ) ; see Fig. 3 for explanation of the t-test outcomes

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6